Low complexity detectors for uplink massive MIMO based on a refinement of linear algorithms and efficient initialization

Massive multiple-input multiple-output (mMIMO) plays a crucial role in improving the quality-of-service and achieving high power efficiency and spectrum efficiency in beyond fifth generation communication systems. However, data detection in uplink mMIMO is not a trivial task as the computational complexity increases with the number of antennas. The equalization matrix is diagonally dominant, and hence, most of the existing linear detectors use the diagonal matrix. Unfortunately, detection based on a diagonal matrix may require a high number of iterations to converge, which increases the computational complexity. This is highly challenging because of the large number of antennas on both the transmitting and receiving sides. In this paper, we propose a refinement of six linear mMIMO detectors based on a band matrix formulation to accelerate the convergence rate, and hence reduce the complexity. The proposed linear detectors include the Newton iterations method, the Neumann series method, the accelerated over-relaxation method, the successive over-relaxation method, the Gauss-Seidel (GS) method, and the Jacobi method. The computation of the band matrix inverse is also presented in this paper and employed in the proposed detectors. In addition, efficient initialization based on the structure of the band matrix is proposed, which both improves the convergence rate and yields a substantial performance gain. Simulations show that the proposed detectors achieve minimum mean-squared error performance with significant complexity reduction even when the number of users approaches the number of base station antennas. It is also shown that the refined detector based on the GS and band matrix achieves the highest performance gain with the lowest computational complexity.